The toolkit's strategy is simple: build every surface on the one shared engine. New interfaces are thin adapters, so they inherit the same read-only, offline-first, deterministic behaviour for free. See Architecture.
- CLI (
devops-ai) — analyze, explain, validate, list, serve, version. See CLI guide. - Python SDK —
AnalysisEngineand the public models. See SDK guide. - REST API — FastAPI with Swagger/ReDoc. See REST API guide.
- Extensibility — YAML signatures and pluggable AI providers via
register_provider(). See Plugin guide. - Providers — Anthropic, OpenAI, Gemini, Ollama, plus the offline default.
These are design directions, not commitments. Each reuses the engine — none reimplements analysis.
| Surface | Idea | Design note |
|---|---|---|
| Web UI | Browser front-end over the REST API for paste-and-analyze + browsing | web-ui.md |
| VS Code extension | Analyze the active file / selection / terminal output inline | vscode-extension.md |
| GitHub Action | Analyze CI logs and manifests in pipelines; annotate PRs | github-action.md |
| MCP server | Expose analyze/explain/validate as tools to AI agents | mcp-server.md |
| Desktop app | Offline, local desktop wrapper for operators | — |
- Broader knowledge base — more signatures across the supported technologies and new ones. Contributions welcome; it's just YAML. See Knowledge base.
- Smarter detection and ranking — improved technology/source-kind detection and scoring.
- More providers — additional adapters as the ecosystem evolves.
- Read-only. No surface will ever execute commands or mutate infrastructure. See Security.
- Offline-first & deterministic. The core never requires an API key; LLMs stay optional.
- One engine. Every interface is a thin adapter over
AnalysisEngine. - Additive contracts.
AnalysisResultgrows by addition so interfaces never drift.
- Propose or build a surface — open an issue or PR. See Contributing.
- Follow progress: https://devopsaitoolkit.com/blog and https://devopsaitoolkit.com/newsletter.